AI, like so many other new technologies, has generated too many unrealistic expectations. Too many organizations today liberally sprinkle references to neural networks, machine learning and other forms of the technology on their sites with little link to its actual capabilities. Merely calling a site “AI-powered,” other than maybe helping with fundraising, doesn’t make it any more effective.
Artificial Intelligence has been around for longer than we realize. From our forefathers who dreamt of mechanical men designed to help them with their labor to early computers conceived as logical machines, we’ve already come a long way. Engineers have always been attempting to create mechanical brains by outsourcing a brains work by reproducing capabilities such as basic arithmetic and memory. But the real change happens when rather than teaching computers everything they need to know to carry out tasks, we attempt to teach them to learn for themselves. Today, with internet and the massive increase in the amount of digital data being stored, generated, and made available for analysis, it is far more efficient to code machines to think like humans, and then connect them to the internet to give them access to all of the data in the world.
Where do Machine Learning, Neural Networks, and NLP fit in?
Artificial Intelligence systems are usually classified as either applied or general. Far more common is Applied AI. These systems are designed to intelligently maneuver an autonomous vehicle or trade shares and stocks and more. Far less common are generalized AI applications which can theoretically do any job, but this is where a whole lot of exciting advancements are happening. This is also the area that has proved instrumental in the development of Machine Learning.
Machine Learning applications can listen to music, decide whether it can make a certain user happy or sad, and find other music to imbibe the same mood. These applications can read the text and work out whether the person who wrote it is making a complaint or offering congratulations. In some instances, they could even compose music on their own along the same themes, or identify the same type of songs that will be liked by the admirers of the original song. These are all possibilities offered by systems based on ML and neural networks.
The major advancement in teaching computers to decipher the world like us, while retaining the inherent advantages they have over us like speed, accuracy, and lack of bias, has been possible thanks to the development of neural networks. Neural Networks work in the same way a human brain does, by classifying information. It can be taught to recognize images for instance and classify them based on the elements they contain. It primarily functions on a system of probability, based on data fed to it. It can decide and make predictions with a degree of certainty. With the addition of a feedback loop to the process, we help it “learn.” It changes the approach it takes in the future, everytime it's told whether its decisions are correct or incorrect. To be able to interact and communicate with devices as naturally as we would with other humans, we need Natural Language Processing (NLP). In recent years, this field has turned into a source of immensely path-breaking innovation, and one which relies heavily on ML.
What can A.I. do today?
Let’s start with what AI is already doing. The biggest advances have been cognition and perception. In the field of perception, speech is where some of the most practical advances have been made. Voice recognition is far from perfect, but many people are now using it, think Alexa, Siri, and Google Assistant. Speech recognition is now three times faster on average than typing on a cell phone. This substantial improvement has come since the summer of 2016, not over the past ten years.
Image recognition has improved dramatically too. Apps like Facebook now recognize many of your friends’ faces in photos posted and prompt you to tag them with their names. An app running on your smartphone can recognize just about any bird in the wild. Image recognition is replacing ID cards at corporate headquarters. Back in the day, vision systems, like the ones used in self-driving cars, made a mistake identifying a pedestrian once in 30 frames, the cameras recorded 30 frames a second. Today, they err less than once in 30 million frames. Voice recognition even in noisy environments is now nearly equal to human performance.
In the space of cognition and problem solving too, machines have beaten the finest human players of Go and poker. Experts predicted that these achievements would take at least another decade. But we’re there already. Intelligent agents are being used to detect malware and to prevent money laundering. An insurance company in Singapore is using IBM technology to automate the claims process and a data science platform firm uses a system from Lumidatum to offer timely advice to improve customer support. A substantial amount of businesses are using ML systems to determine what trades they will execute on Wall Street, and an increasing number of credit decisions are being made with its assistance. To better the recommendations of products to users and optimize inventory, Amazon is employing ML.
At many tasks that were once best done by humans, Machine learning systems are now superior. Although these systems are far from perfect, their error rate is better than human-level performance. This has opened up vast new possibilities for transforming the economy and the workplace. Every time an AI-based system surpasses human performance at a given task, they spread faster.